14
14
# You should have received a copy of the GNU General Public License
15
15
# along with this program; if not, write to the Free Software
16
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
16
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
18
18
"""Bzrlib specific gzip tunings. We plan to feed these to the upstream gzip."""
20
from __future__ import absolute_import
22
from cStringIO import StringIO
20
24
# make GzipFile faster:
22
from gzip import U32, LOWU32, FEXTRA, FCOMMENT, FNAME, FHCRC
26
from gzip import FEXTRA, FCOMMENT, FNAME, FHCRC
27
__all__ = ["GzipFile"]
31
# we want a \n preserved, break on \n only splitlines.
32
from bzrlib import symbol_versioning
34
__all__ = ["GzipFile", "bytes_to_gzip"]
38
"""Return i as an unsigned integer, assuming it fits in 32 bits.
40
If it's >= 2GB when viewed as a 32-bit unsigned int, return a long.
48
"""Return the low-order 32 bits of an int, as a non-negative int."""
49
return i & 0xFFFFFFFFL
52
def bytes_to_gzip(bytes, factory=zlib.compressobj,
53
level=zlib.Z_DEFAULT_COMPRESSION, method=zlib.DEFLATED,
54
width=-zlib.MAX_WBITS, mem=zlib.DEF_MEM_LEVEL,
56
"""Create a gzip file containing bytes and return its content."""
57
return chunks_to_gzip([bytes])
60
def chunks_to_gzip(chunks, factory=zlib.compressobj,
61
level=zlib.Z_DEFAULT_COMPRESSION, method=zlib.DEFLATED,
62
width=-zlib.MAX_WBITS, mem=zlib.DEF_MEM_LEVEL,
64
"""Create a gzip file containing chunks and return its content.
66
:param chunks: An iterable of strings. Each string can have arbitrary
70
'\037\213' # self.fileobj.write('\037\213') # magic header
71
'\010' # self.fileobj.write('\010') # compression method
72
# fname = self.filename[:-3]
76
'\x00' # self.fileobj.write(chr(flags))
77
'\0\0\0\0' # write32u(self.fileobj, long(time.time()))
78
'\002' # self.fileobj.write('\002')
79
'\377' # self.fileobj.write('\377')
81
'' # self.fileobj.write(fname + '\000')
83
# using a compressobj avoids a small header and trailer that the compress()
84
# utility function adds.
85
compress = factory(level, method, width, mem, 0)
89
crc = crc32(chunk, crc)
90
total_len += len(chunk)
91
zbytes = compress.compress(chunk)
94
result.append(compress.flush())
95
# size may exceed 2GB, or even 4GB
96
result.append(struct.pack("<LL", LOWU32(crc), LOWU32(total_len)))
97
return ''.join(result)
30
100
class GzipFile(gzip.GzipFile):
50
120
Yes, its only 1.6 seconds, but they add up.
123
def __init__(self, *args, **kwargs):
124
symbol_versioning.warn(
125
symbol_versioning.deprecated_in((2, 3, 0))
126
% 'bzrlib.tuned_gzip.GzipFile',
127
DeprecationWarning, stacklevel=2)
128
gzip.GzipFile.__init__(self, *args, **kwargs)
53
130
def _add_read_data(self, data):
54
131
# 4169 calls in 183
55
132
# temp var for len(data) and switch to +='s.
60
137
self.extrasize += len_data
61
138
self.size += len_data
140
def _write_gzip_header(self):
141
"""A tuned version of gzip._write_gzip_header
143
We have some extra constrains that plain Gzip does not.
144
1) We want to write the whole blob at once. rather than multiple
145
calls to fileobj.write().
146
2) We never have a filename
147
3) We don't care about the time
150
'\037\213' # self.fileobj.write('\037\213') # magic header
151
'\010' # self.fileobj.write('\010') # compression method
152
# fname = self.filename[:-3]
156
'\x00' # self.fileobj.write(chr(flags))
157
'\0\0\0\0' # write32u(self.fileobj, long(time.time()))
158
'\002' # self.fileobj.write('\002')
159
'\377' # self.fileobj.write('\377')
161
'' # self.fileobj.write(fname + '\000')
63
164
def _read(self, size=1024):
64
165
# various optimisations:
65
# reduces lsprof count from 2500 to
166
# reduces lsprof count from 2500 to
66
167
# 8337 calls in 1272, 365 internal
67
168
if self.fileobj is None:
68
169
raise EOFError, "Reached EOF"
109
212
# (The number of bytes to seek back is the length of the unused
110
213
# data, minus 8 because those 8 bytes are part of this member.
111
214
seek_length = len (self.decompress.unused_data) - 8
113
assert seek_length > 0
216
# we read too much data
114
217
self.fileobj.seek(-seek_length, 1)
218
self._gzip_tail = self.decompress.unused_data[0:8]
219
elif seek_length < 0:
220
# we haven't read enough to check the checksum.
221
if not (-8 < seek_length):
222
raise AssertionError("too great a seek")
223
buf = self.fileobj.read(-seek_length)
224
self._gzip_tail = self.decompress.unused_data + buf
226
self._gzip_tail = self.decompress.unused_data
116
228
# Check the CRC and file size, and set the flag so we read
117
229
# a new member on the next call
122
234
"""tuned to reduce function calls and eliminate file seeking:
124
236
reduces lsprof count from 800 to 288
126
238
avoid U32 call by using struct format L
129
# We've read to the end of the file, so we should have 8 bytes of
130
# unused data in the decompressor. If we dont, there is a corrupt file.
241
# We've read to the end of the file, so we should have 8 bytes of
242
# unused data in the decompressor. If we don't, there is a corrupt file.
131
243
# We use these 8 bytes to calculate the CRC and the recorded file size.
132
244
# We then check the that the computed CRC and size of the
133
245
# uncompressed data matches the stored values. Note that the size
134
246
# stored is the true file size mod 2**32.
135
crc32, isize = struct.unpack("<LL", self.decompress.unused_data[0:8])
247
if not (len(self._gzip_tail) == 8):
248
raise AssertionError("gzip trailer is incorrect length.")
249
crc32, isize = struct.unpack("<LL", self._gzip_tail)
136
250
# note that isize is unsigned - it can exceed 2GB
137
251
if crc32 != U32(self.crc):
138
raise IOError, "CRC check failed"
252
raise IOError, "CRC check failed %d %d" % (crc32, U32(self.crc))
139
253
elif isize != LOWU32(self.size):
140
254
raise IOError, "Incorrect length of data produced"
142
256
def _read_gzip_header(self, bytes=None):
143
257
"""Supply bytes if the minimum header size is already read.
145
259
:param bytes: 10 bytes of header data.
147
261
"""starting cost: 300 in 3998
243
357
# 4168 calls in 417.
244
358
# Negative numbers result in reading all the lines
247
content = self.read(sizehint)
248
return content.splitlines(True)
360
# python's gzip routine uses sizehint. This is a more efficient way
361
# than python uses to honor it. But it is even more efficient to
362
# just read the entire thing and use cStringIO to split into lines.
365
# content = self.read(sizehint)
366
# return bzrlib.osutils.split_lines(content)
367
content = StringIO(self.read(-1))
368
return content.readlines()
250
370
def _unread(self, buf, len_buf=None):
251
371
"""tuned to remove unneeded len calls.
253
373
because this is such an inner routine in readline, and readline is
254
374
in many inner loops, this has been inlined into readline().
256
376
The len_buf parameter combined with the reduction in len calls dropped
257
the lsprof ms count for this routine on my test data from 800 to 200 -
377
the lsprof ms count for this routine on my test data from 800 to 200 -
260
380
if len_buf is None:
278
398
self.offset += data_len
280
400
def writelines(self, lines):
281
# profiling indicated a significant overhead
401
# profiling indicated a significant overhead
282
402
# calling write for each line.
283
403
# this batch call is a lot faster :).
284
404
# (4 seconds to 1 seconds for the sample upgrades I was testing).
285
405
self.write(''.join(lines))
407
if sys.version_info > (2, 7):
408
# As of Python 2.7 the crc32 must be positive when close is called
410
if self.fileobj is None:
412
if self.mode == gzip.WRITE:
413
self.crc &= 0xFFFFFFFFL
414
gzip.GzipFile.close(self)